Exposure to Opposing Views can Increase Political Polarization

Chris Bail
Duke University

paper: https://osf.io/preprints/socarxiv/4ygux
website: https://www.chrisbail.net
github: https://github.com/cbail

My collaborators

Political Polarization in the U.S.

Echo Chambers on Facebook

Echo Chambers on Twitter

How does exposure to opposing views shape political polarization?

Hypothesis #1: Intergroup Contact

Hypothesis #2: Backfire Effects

Hypothesis #3: Assymetric Polarization

Pre-Registered Hypotheses

1) Intergroup contact will reduce political polarization

Pre-Registered Hypotheses

1) Intergroup contact will reduce political polarization

2) Backfire effects will increase political polarization

Pre-Registered Hypotheses

1) Intergroup contact will reduce political polarization

2) Backfire effects will increase political polarization

3) Conservatives will be more likely to exhibit backfire effects than liberals

Research Design

Eligibility Criteria

1) Must be living in the United States

Eligibility Criteria

1) Must be living in the United States

2) Must visit Twitter at least three times per week

Eligibility Criteria

1) Must be living in the United States

2) Must visit Twitter at least three times per week

3) Must describe themselves as either a Republican or a Democrat

Outcome: Ideological Consistency Scale

1) “Stricter environmental laws and regulations cost too many jobs and hurt the economy.”

2) “Government regulation of business is necessary to protect the public interest.”

3) “Poor people today have it easy because they can get government benefits without doing anything in return.”

4) “Immigrants today strengthen our country because of their hard work and talents.”

5) “Government is almost always wasteful and inefficient.”

Outcome: Ideological Consistency Scale

6) “The best way to ensure peace is through military strength.”

7) “Racial discrimination is the main reason why many black people can't get ahead these days.”

8) “The government today can't afford to do much more to help the needy.”

9) “Business corporations make too much profit.”

10) “Homosexuality should be accepted by society.”

Treatment: Twitter Bots

Measuring Treatment Compliance

The Cutest Animals on the Internet*

*according to my daughter

Substantive Compliance Measure

“Over the past three days, the [name of study's Twitter account here] retweeted a message about a philanthropist who gave a large amount of money to help people recover from a major disaster. How much money did this person donate?”

Control Variables

1) Frequency of Twitter use

Control Variables

1) Frequency of Twitter use

2) Strength of partisanship

Control Variables

1) Frequency of Twitter use

2) Strength of partisanship

3) Interest in current events

Control Variables

1) Frequency of Twitter use

2) Strength of partisanship

3) Interest in current events

4) Ideological homophily (online)

Control Variables

1) Frequency of Twitter use

2) Strength of partisanship

3) Interest in current events

4) Ideological homophily (online)

5) Ideological homophily (offline)

Control Variables

1) Frequency of Twitter use

2) Strength of partisanship

3) Interest in current events

4) Ideological homophily (online)

5) Ideological homophily (offline)

6) Demographics/SES (age, gender, income, education, region)

Control Variables

1) Frequency of Twitter use

2) Strength of partisanship

3) Interest in current events

4) Ideological homophily (online)

5) Ideological homophily (offline)

6) Demographics/SES (age, gender, income, education, region)

7) Many others…

Initial Response Rate: 42.7%

Addressing Causal Interference

Identify Verification

Block Randomization

Compliance

Attrition

Results

Change in Ideology after 1 Month

Mechanisms?

Robustness Checks

Compound Treatment

Partisan Learning?

Attrition Bias

Heterogeneous Treatment Effects

Extremist Effects

Experiment Effects

Limitations

Non-Random Sample

Representativeness

External Validity/Incentives

Elite Effects?

What is the Mechanism?

Contributions

1) First study to experimentally disrupt social media echo chambers in order to examine how they influence political views.

Contributions

1) First study to experimentally disrupt social media echo chambers in order to examine how they influence political views.

2) No evidence that inter-group contact reduces political polarization, but new evidence of backfire effects—and assymetric polarization.

Contributions

1) First study to experimentally disrupt social media echo chambers in order to examine how they influence political views.

2) No evidence that inter-group contact reduces political polarization, but new evidence of backfire effects—and assymetric polarization.

3) A new model for research that combines social media data, survey data, network analysis, text analysis, and artificial intelligence—and provides new techniques for identifying causal interference and demographic disimulation in survey research.